Featured
Experience
Technologies
• Worked across 9+ AI projects, leading 2 projects end-to-end across financial message validation, enterprise search, multilingual customer support, document intelligence, AI agents, travel automation, and domain-specific LLM applications.
• Led key development for KAI BOT, an enterprise AI assistant for the R Systems Sales team, working across 15,000+ heterogeneous files to automate document discovery, information retrieval, and sales-content generation, including automated pitch-deck composition and retrieval of relevant sales material using RAG, vector search, OCR, and metadata-driven pipelines.
• Developed and contributed to multiple production-oriented AI systems involving RAG, NLP, LLM orchestration, agentic workflows, document processing, and fine-tuned models, translating business workflows into practical AI-powered applications across different enterprise use cases.
• Led the development of AI-powered MT/MX financial message validation systems, building workflows to understand and validate structured financial messages, identify errors and inconsistencies, and automate parts of the validation process, with continued ownership into the system’s second phase.
Featured
Projects
No projects to show
Projects will appear here once published. In the meantime, visit the blog or head home.
About
Me
Vansh Garg
I focus on building production-grade AI systems that move beyond demos and deliver real operational impact. My work centers on RAG pipelines, LLM-powered automation, and agentic workflows that reduce manual work, improve decision-making, and scale reliably in production environments.
Skills
Featured
Blogs
Your AI Agent Needs a Kill Switch (Most Don't Have One)
An OpenAI agent hacked Hugging Face over five days and nobody could stop it mid-run. Here is why every production agent needs a way to be shut off, and how to build one.
The Layer Every AI Agent Is Missing: Guardrails and Intent Detection
I shipped an agent that tried to email a salary database because it never checked what the user actually wanted. Here is the layer I should have built first.
Hey, you scrolled this far, let's talk.
Development